dsh-dream: A Plugin for Organizing and Reusing Task Experience
dsh-dream is a DeepSeek Harness plugin designed to address the issues of models forgetting context and repeatedly making the same mistakes. It helps users build long-term memory through automated extraction, structured storage, and a review mechanism. The plugin supports browsing and retrieval, automatic reuse of adopted experiences, manual review of dream diaries, and writing experience rules into documentation. After completing specific tasks, users can trigger automatic organization; once reviewed and adopted, related new tasks will automatically reuse these experiences, enabling the accumulation and reuse of agent capabilities.
Read Moredsh-dream: DSH Memory Backend Integration Plugin
`dsh-dream` is a standalone memory consolidation plugin in the DeepSeek Harness ecosystem, designed to address long-term memory fragmentation and redundancy. It periodically deduplicates, updates, and prunes recent conversations and existing memories via background silent consolidation (by default every 5 conversations or 24 hours) and manual triggering (the `/dream` command). The plugin supports atomic overwrites and historical snapshots, offers flexible configuration, and effectively maintains the cleanliness and accuracy of the memory store.
Read More